Photo by writer | Ideogram
Introduction
Large models of language You have transformed all the prices of national intelligence over recent years, marks the beginning of the new season in Ai's history. Often called their llms, it turned their communication mechanisms, whether we return the information, ask questions, or to produce different human language content.
As the llms also deals with our daily life and specialists, it is very important to understand the ideas and basics around them, both construction and practical use and application.
In this article, we examine the big words of ten languages that are key languages of understanding these powerful AI systems.
1. a transformer of construction
Definition: Transformer is the basis for large-language models. It is a deep Neareral Nealural invention to its highest expothent, including various layers and layers such as feeding networks and your attention allowing a simple processes and consecutive insertion.
Why is it important: Due to transformer construction, an understanding of complex languages and produces the language effects at an unprecedented scale, overcome the limitations of solutions of the prior environment.
2. Attention to pay attention
DefinitionInitially contemplated in language translation activities in normal neural networks, various lengths and various difficulties and various difficulties and various difficulties and various difficulties and various difficulties and various difficulties and various difficulties and difficulties and difficulties and difficulties Different variety and various difficulties and various difficulties and difficulties, and various difficulties and various difficulties and various difficulties and various difficulties and various difficulties and various difficulties and various difficulties and various difficulties. While the basic path is not like part of the transformer's lower transformer buildings, they set the foundations of improved methods (as we can discuss soon).
Why is it important: Techniques are essential to synchronization and alternative texture in tasks such as translating and summarizing, modifying the learning processes and generators that are most effective in content activities.
3. Satisfaction
Definition: If there is a type of transformer construction of the transformation of the success of the prosperity of the LLMS, that is a way of being attentive. Self-control overruns the limitations of regular attention as a subsequent workforce by allowing each word – or token, respectively – token) at the same time, no matter what.
Why is it important: Attention to depend on, patterns, and the combination between the things in the same order is most useful for the deeper release and renewable login order, and the target order produced as a compatible response and content.
4. Encoder and Decoder
Definition: The Classical Transformmer building is divided into two main parts or halves: encoder and decoder. The Encoder has an obligation to process and seize the installation of the Map Deeper Map, and Decoder focuses on the production of step-action-step measures using both previously produced sections and the leading components of the Encoder. Both parties are linked, so the decoder has received processing results from the Encoder (called Hidn States) as installation. In addition, both Encoder and the internal decoder “are repeated” in the form of many Encoder and Decoders, respectively: This Standard of depth helps the model learn the remaining features and output.
Why is it important: Encoder combination and decoder, each has its own self-style, is the key to measuring an understanding of the input of the LLM generation.
5. FIRST TRANSPORT
Definition: Like the bases of the house from the beginning, previous training is a Co-train training for the first time, that is, gradually reading all the parameters of their models or metals. The size of these models have taken billions of parameters. Therefore, pre-training is a very expensive process lasting to weeks to complete and requires a large and varied text company.
Why is it important: Previous training is important to create an unknown LLM and install the general language patterns and semantics in all headquators.
6. Good organization
Definition: In contrasting pre-training training, good formation is a process of taking a previously trained LLM and training and a small domain set for examples. While costly, good order is not more expensive than pre-model training from the beginning, and often installs the costumes of only clothes only for buildings.
Why is it important: Having a focus on the visual concrete and app centers such as legal analysis, medical diagnosis, or customer support is important because customer-trained models, special names, and compliance requirements.
7. Empowerment
Definition: The equipment and models of AI really don't really understand the language, but just numbers. This applies and in llms, so while we are typically talking about the models “straight and generating the language”, what they do to manage important numbers of the wrong language
Why is it important: The order of the map text in the embarking incidents make it possible to make llms to make the same thinking, and the usual of the data in the other side, all without losing advanced structures; Thus, the green answers made of model can be returned to a solid language that we agree with the relevant information.
8. DMPT Engineering
Definition: The end of the llMS users should be familiar with good use of these models for achieving their goals, and instant engineering has shown as techniques and active until it is valid. Defect Engineering includes a set of guidelines and designs to design practical answers that guide the model to produce accurate, accurate, guidance.
Why is it importantEVERITAGES, OVERIAL, OVERIAL, and idealized LLM results especially is a matter of learning high-quality, specified, clarified, and converted question.
9. reading the content
Definition: It is also called Shot-Shot Learning, this is a way of teaching the llms to perform new jobs paid forward in providing for examples of desired results, without a model or model. It may be considered a special engineering system, as completely obtained by the information received by the model during pre-training.
Why is it important: The context is included as an effective and effective way to learn to solve new jobs according to examples.
10. Parameter Country
Definition: The size of the LLM difficulty is usually measured by several factors, the parameter calculation to be one of them. Names of known models are like GPT-3 (172-parameter) and Llama-2 (with up to 70b) show clearly the importance of measuring skills and llm exposure to language production. The number of parameters is important when it comes to the llm's skills, but some features such as the number and quality of training data, construction, and the good planning methods used are also similarly used.
Why is it important: The parameter figure is not only in define the 'last' volume and measured its functionality in consultation activities and storage measures, especially when installing various detailed conversations between user and model.
Rolling up
The document explored the importance of ten words around the major languages: main focus on the attention of the entire AI, due to incredible success in the past few years. A coupled with these concepts put you in a good position to stay familiar with the new tendency and development in the formation of the fastest of the fastest.
Iván Palomares Carrascus He is a leader, writer, and a counselor in Ai, a machine study, a deep reading and llms. He trains and guides others to integrate AI in the real world.
